Lung parenchyma segmentation method based on convex hulls correcting by parabolic

Abstract

The invention relates to a lung parenchyma segmentation method based on convex hulls correcting by parabolic. The lung parenchyma segmentation method includes extracting lung parenchyma contour images, restoring lung parenchyma external contour and restoring lung parenchyma internal edges sequentially. The lung parenchyma segmentation method is characterized in that lung parenchyma external contour restoration is realized through the steps of 1, ranking points in lung parenchyma contour image edge contour; 2, establishing edge stacks of points in step one, and obtaining an edge point set P= {p0, p1, ... pn} (n> / =3); 3, connecting two adjacent points in the point set P to be segments, ranking the segments in descending order of length, and finding segments pdpd +1 of pulmonary nodules corresponding to missing parts of lobe contour according to segment length and position; 4, utilizing the parabolic to correct the segments pdpd +1, and obtaining lung parenchyma edge contour.

Description

technical field [0001] The invention relates to the field of medical image processing, in particular to a lung parenchyma segmentation method based on a parabola modified convex hull. Background technique [0002] With the development of computer tomography (CT) technology and computer image processing technology, the computer-aided diagnosis system (Computer-Aided Detection) provides a new means for accurate, fast and automatic analysis and processing of CT image data. Help to improve the sensitivity and specificity of doctors' diagnosis. In the study of computer-aided diagnosis of lung diseases, the segmentation of lung parenchyma can eliminate the influence of irrelevant factors in CT images (such as chest, heart, examination bed, etc.), reduce unnecessary calculations, and the accuracy of segmentation is the most important factor. Quantitative analysis of lung disease, detection of pulmonary nodules and key link in lung function assessment. [0003] In CT images, since...

AI Technical Summary

Problems solved by technology

The existing correction algorithms mainly include: the rolling ball algorithm proposed by Armato S.G., etc., this method uses a ball with a radius R to roll along a certain direction on the edge of the extracted lung area, and judges whether the points in the ball are mutually compatible Eight connected domains are used to judge whether the edge needs to be repaired, but the radius R of the sphere will have a great impact on the repair result. At the same time, the selection of the R value is usually based on experience, which reduces the applicability of the algorithm and often causes overcompensation. Phenomenon

The mathematical morphology method proposed by Hu S. et al. segmented the lung parenchyma, which has high computational complexity and will affect the accuracy of the edge

Bollotti R. et al. selected the Glued Elastic Band (GEB) algorithm to repair the lung boundary, imitating the mechanical principle to include the concave part with a smaller bending radius in the required area, while the part with a larger bending radius is still excluded. In addition, this method is computationally complex, and the repair result is also affected by the sampling step size

In addition, Kanazawa K et al. proposed a curvature-based method. According to the large curvature change in the missing part of the lung parenchyma, the curvature threshold is set and the missing part is corrected, but the curvature on the lung boundary is easily affected by noise. impact, and a large amount of calculation

Wang Jing uses the method of endpoint detection, but only for nodules that are not too small or too large (3mm)

Yuan Kehong et al proposed to correct the original lung contour based on the calculation of the local convex hull, and re-include the lung nodules that were originally excluded by mistake, which can effectively correct the initial edge error caused by high-density nodules and blood vessels in the lung, but Due to the need to set multiple parameters, the automatic adaptability to the nodule size is poor, and due to the characteristics of the convex hull algorithm, it is not suitable for the detection of nodules on the inner edge of the lung lobe (there are many natural physiological depressions on the inner edge, such as blood vessels) , heart), it is difficult to obtain a good segmentation effect, and the result of repairing with a straight line cannot completely include the nodules inside the edge

Images